Scenario

You are designing a customer-facing experiment for a workflow change in a professional software product. The feature can affect repeated usage across visits, devices, and collaborators, so you need to decide whether to randomize by user, session, device, account, or another unit.

Question

How would you choose the unit of randomization for this experiment? Walk through the trade-offs you would consider, how the choice affects measurement and validity, and how you would validate that the final design is trustworthy.